09_subj-ratings_analysis

Check control responses for filtered data

Additional Info

Comparing Ratings of the Seed Stories

• How compelling do you find the evidence? The condition manipulation was least effective in the professor seed and most effective in smuggler.
• The professor seed is rated most emotional
• THe evidence distribution looks similar to suspect_committedCrime. The same hold for suspect_conviction and suspect_convictionJustified.

Comparison of Conditions (generalized over all generations)

• Questions that are condition-sensitive: author_belief, evidence, suspect_committedCrime, suspect_conviction, suspect_convictionJustified
• Questions that are not condition-sensitive: author_trust (very small), reader_emotion, story_subjectivity

Judgments over generations and conditions

• Conditions that have the same direction of development over generations:
• author_trust (decreases)
• reader_emotion (decreases – a little)

• story_subjectivity (stable; no effects over generations)

• Strong evidence condition always decreases over generations, except for story_subjectivity where it is stable

• Weak evidence condition is generally stable or increases, except for author_trust and reader_emotion

Interesting Observation

• Weak condition: certainty that the suspect committed the crime increases, whereas suspect_conviction stays stable –> in fifth generation, big difference between suspect_committedCrime and suspect_conviction (this effect also seems to exist in the strong condition but much less)
• Interesting correlation also to author_belief and evidence (higher uncertainty about evidence doesn’t get reflected in certainty about suspect’s guilt; author_belief shows similar distribution as evidence and suspect_committedCrime)

By Story Split

• We generally see that the suspect becomes more guilty over time in the weak condition. We don’t see that in the professor and scam story.

Professor seed gender differences

• Gender effect in overall suspect_conviction and suspect_convictionJustified ratings. In- vs. out-group-effects (consider that evidence manipulation was least accpeted in this story.)

Correlation between questions

• Highest correlation in weak and strong condition: suspect_committedCrime – suspect_convictionJustified
• Followed by: suspect_committedCrime – evidence
• Interesting difference between conditions: suspect_committedCrime – author_belief (0.52 for weak and 0.35 for strong)

Split by condition

0.5 tendency

• Movements away from 0.5: author_belief weak, maybe evidence weak, reader_emotion strong and weak, suspect_committedCrime weak, suspect_convictionJustified weak

• the variance stays fairly stable over generations

Jaccard distance

• Jaccard distance increases the higher the generation, but all means are very high

Heges

• The use of hedges decreases over generations (what does it look like in a ratio with total length?)
• weak and strong condition look almost identical

Link between hedges and guiltiness of the suspect

There seems to be a small difference between conditions.